Determining the optimal K from the bNMF run summary
The plot “Convergence Scores by K” helps analyze which K value
performs better in terms of convergence.


Determining the optimal cutoff for clustering weights (parameters:
N, M)
1) Fit a 1st line to the top N% of weights
2) Fit a 2nd line to the M% of tail weight
3) Using the remaining weights from top N% to last M%, check if they
have shorter distance to 1st or 2nd line
4) The first weight that has a shorter distance to 2nd line (defined
by long tail) is selected as the cutoff.

Optimal weight cutoff: 0.59778 (includes top 8.1% of variants)
Cluster Weights
Weights above cutoff are highlighted
Manhattan plot

Cluster Weight Heatmaps
Column label color correspond to the cluster with the column’s
highest weight
Genes x Clusters

Variants x Clusters

Cluster Circle Plots
Only includes variants and phenotypes with weights above cutoff
Blue = negative trait
Red = positive trait
Green = variant